Oncofusion

start / graph / ana-q130-the-four-mark-panel-is-two-active-marks-down-and-two-repressive-marks-up-in-the-same-1228-bins

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q130-does-the-eviction-gradient-exist-in-h3k79me2

With H3K79me2 added, the eviction-depth panel over one set of 1,228 bins under one drug is H3K27me3 +1.000, H2AK119ub +1.000, H3K4me3 -1.000 and H3K79me2 -1.000 - a clean repressive-up / active-down split that no shared library, blacklist or regression artefact can produce

also called the four-mark panel · two up two down · the free artefact control, doubled

Four different chemical tags were measured on the same stretches of DNA in the same cells under the same drug. The two 'off' tags go up where the fusion is most removed and the two 'on' tags go down there. A technical glitch in the measurement could not push two tags one way and two the other.

The same thing, for a scientist

**All five contrasts below are the same 1,228 bins, the same four eviction-depth quartiles, the same murine MLL-AF9 culture, the same 500 nM VTP-50469 48 h against DMSO 72 h, and the same within-sample-share normalisation.** Three were computed by Q125 in cycle 88 and are read from its committed JSON; two are this beat's [E1][E2]. | mark | Q1 | Q2 | Q3 | Q4 | Spearman vs eviction depth | span | aggregate | |---|---:|---:|---:|---:|---:|---:|---:| | H3K27me3 | 2.0007 | 1.5120 | 1.1304 | 0.9780 | **+1.000** | 1.0227 | 1.3612 | | H2AK119ub `paired` | 1.4450 | 1.1483 | 1.1238 | 1.0630 | **+1.000** | 0.3820 | 1.1694 | | H2AK119ub `rep2` | 1.0151 | 0.9063 | 0.8878 | 0.8584 | **+1.000** | 0.1567 | 0.9071 | | H3K4me3 | 0.9491 | 1.1298 | 1.1761 | 1.2025 | **-1.000** | -0.2534 | 1.1276 | | **H3K79me2** | **0.5860** | **0.8188** | **0.9280** | **1.1865** | **-1.000** | **-0.6005** | 0.8156 | **Q125's control was two marks at +1.000 against one at -1.000. It is now two against two**, and the second member of the active pair is the mark with the largest span of the four [E1][E2]. **Why this is the load-bearing control and not decoration.** A regression-to-the-mean effect, a blacklist gap, a mappability feature of the fusion's strongest sites, a copy-number difference between the quartiles - every mechanism of that family acts on the BINS and must therefore act the same way on every track measured in those bins. **None can reverse sign between marks.** The matched null confirms the mechanism is available: all three inputs carry their own monotone gradient across the same quartiles at Spearman -1.000, so the bins genuinely are not interchangeable - and the mark ratios still split two and two [E3]. **What the panel cannot rule out** is a confound that acts on ANTIBODY CLASS rather than on bins - for instance a systematic difference in how repressive-mark and active-mark ChIPs behave at high-occupancy sites under any perturbation. No control in this deposit addresses that, and it is recorded as the residual.

Go deeper
The ordering within each pair is informative and the panel is not symmetric

The two repressive marks differ by a factor of 2.7 in span (H3K27me3 1.0227 against H2AK119ub paired 0.3820) and the two active marks by 2.4 (H3K79me2 0.6005 against H3K4me3 0.2534). The largest and the second-largest spans belong to marks in OPPOSITE classes, which is what a genuine chromatin switch looks like and is not what a class-level antibody artefact would look like - the latter would predict the two members of each class to move together in magnitude as well as sign.

The two H2AK119ub conventions are quoted rather than averaged because Q125 measured them disagreeing in sign on a different criterion. Their spans differ by 2.4-fold (0.3820 against 0.1567) while agreeing on sign, which is the honest statement of how well a span reproduces across two libraries of one culture in this series - and it is the only such estimate available anywhere in this deposit, because no other mark has two conventions.

What a reader should take from the aggregates column

Read the aggregates alone and the five contrasts say: H3K27me3 up 36%, H2AK119ub up 17% or down 9% depending on convention, H3K4me3 up 13%, H3K79me2 down 18%. Four small-to-moderate effects in three directions with no pattern. Read the quartiles and they are a single coherent switch. That is the generalisable point, and it is now demonstrated on five contrasts rather than three.

Facts and figures
scriptanalyses/q130-does-the-eviction-gradient-exist-in-h3k79me2/run.py
statusrun
result summaryFive contrasts on one bin set under one drug: H3K27me3 +1.000 (span 1.0227), H2AK119ub +1.000 (0.3820 / 0.1567), H3K4me3 -1.000 (0.2534), H3K79me2 -1.000 (0.6005).
absence classmeasured-not-analysed
absence class noteThe fourth mark's row was computable from disk with no download. Q125 named it in its own beat as the highest-value remaining item and wrote 'this is the one that could change cycle 83'.
absence class assigned2026-09-23 (Q130)
epigenetic readoutH3K27me3, H2AK119ub, H3K4me3 and H3K79me2 eviction-depth profiles, one bin set
panel
K4_NT_VTP-1.000K4_NT_VTP: -1K79_NT_VTP-1.000K79_NT_VTP: -1K119_PAIRED_NT_VTP1.000K119_PAIRED_NT_VTP: 1K119_REP2_NT_VTP1.000K119_REP2_NT_VTP: 1K27_NT_VTP1.000K27_NT_VTP: 1
panel knockout genotype
K4_C1_VTP-1.000K4_C1_VTP: -1K79_C1_VTP-1.000K79_C1_VTP: -1K119_REP2_C1_VTP0.400K119_REP2_C1_VTP: 0.4K27_C1_VTP0.800K27_C1_VTP: 0.8K119_PAIRED_C1_VTP1.000K119_PAIRED_C1_VTP: 1
inputs
[
  "this beat's H3K79me2 quartile profile",
  "Q125's committed deviation2-quartiles-and-conventions.json"
]
panel — the numbers behind the chart
{
  "K119_PAIRED_NT_VTP": {
    "R_by_quartile_Q1_to_Q4": [
      1.445,
      1.1483,
      1.1238,
      1.063
    ],
    "spearman_vs_eviction_depth": 1,
    "R_aggregate_S_FLAG": 1.1694,
    "span_Q1_minus_Q4": 0.382,
    "source": "READ from q125 deviation2-quartiles-and-conventions.json, not retyped"
  },
  "K119_REP2_NT_VTP": {
    "R_by_quartile_Q1_to_Q4": [
      1.0151,
      0.9063,
      0.8878,
      0.8584
    ],
    "spearman_vs_eviction_depth": 1,
    "R_aggregate_S_FLAG": 0.9071,
    "span_Q1_minus_Q4": 0.1567,
    "source": "READ from q125 deviation2-quartiles-and-conventions.json, not retyped"
  },
  "K27_NT_VTP": {
    "R_by_quartile_Q1_to_Q4": [
      2.0007,
      1.512,
      1.1304,
      0.978
    ],
    "spearman_vs_eviction_depth": 1,
    "R_aggregate_S_FLAG": 1.3612,
    "span_Q1_minus_Q4": 1.0227,
    "source": "READ from q125 deviation2-quartiles-and-conventions.json, not retyped"
  },
  "K4_NT_VTP": {
    "R_by_quartile_Q1_to_Q4": [
      0.9491,
      1.1298,
      1.1761,
      1.2025
    ],
    "spearman_vs_eviction_depth": -1,
    "R_aggregate_S_FLAG": 1.1276,
    "span_Q1_minus_Q4": -0.2534,
    "source": "READ from q125 deviation2-quartiles-and-conventions.json, not retyped"
  },
  "K79_NT_VTP": {
    "R_by_quartile_Q1_to_Q4": [
      0.586,
      0.8188,
      0.928,
      1.1865
    ],
    "spearman_vs_eviction_depth": -1,
    "R_aggregate_S_FLAG": 0.8156,
    "span_Q1_minus_Q4": -0.6005,
    "source": "THIS BEAT"
  }
}
panel knockout genotype — the numbers behind the chart
{
  "K119_PAIRED_C1_VTP": {
    "R_by_quartile_Q1_to_Q4": [
      1.1325,
      1.0404,
      1.0166,
      0.9918
    ],
    "spearman_vs_eviction_depth": 1,
    "R_aggregate_S_FLAG": 1.0379,
    "span_Q1_minus_Q4": 0.1407,
    "source": "READ from q125 deviation2-quartiles-and-conventions.json, not retyped"
  },
  "K119_REP2_C1_VTP": {
    "R_by_quartile_Q1_to_Q4": [
      0.8842,
      0.8809,
      0.8564,
      0.8833
    ],
    "spearman_vs_eviction_depth": 0.4,
    "R_aggregate_S_FLAG": 0.8761,
    "span_Q1_minus_Q4": 0.0009,
    "source": "READ from q125 deviation2-quartiles-and-conventions.json, not retyped"
  },
  "K27_C1_VTP": {
    "R_by_quartile_Q1_to_Q4": [
      1.0134,
      1.0581,
      0.9014,
      0.8058
    ],
    "spearman_vs_eviction_depth": 0.8,
    "R_aggregate_S_FLAG": 0.9334,
    "span_Q1_minus_Q4": 0.2076,
    "source": "READ from q125 deviation2-quartiles-and-conventions.json, not retyped"
  },
  "K4_C1_VTP": {
    "R_by_quartile_Q1_to_Q4": [
      0.7852,
      0.8601,
      0.878,
      0.9259
    ],
    "spearman_vs_eviction_depth": -1,
    "R_aggregate_S_FLAG": 0.8697,
    "span_Q1_minus_Q4": -0.1407,
    "source": "READ from q125 deviation2-quartiles-and-conventions.json, not retyped"
  },
  "K79_C1_VTP": {
    "R_by_quartile_Q1_to_Q4": [
      0.6123,
      0.7955,
      0.8928,
      1.0464
    ],
    "spearman_vs_eviction_depth": -1,
    "R_aggregate_S_FLAG": 0.799,
    "span_Q1_minus_Q4": -0.4341,
    "source": "THIS BEAT"
  }
}
How this could bite us back
  • If the repressive-up / active-down switch is the real pharmacodynamic signature of menin inhibition, then any cell that cannot execute half of it - a polycomb-deficient cell, which is a documented resistance genotype - would still show the active-down half and score as engaged on an H3K79me2 or H3K4me3 assay. The panel makes the resistance blind spot sharper, not smaller.
  • A four-mark panel is a four-antibody assay. Proposing it as a clinical pharmacodynamic readout means proposing four validated ChIPs on a patient sample at two timepoints, which is a real cost and is not obviously deliverable in a marrow aspirate.
Still open
  1. Is there a fifth mark to add? H3K36me2/3 and H3K9me3 are not deposited in the NTvKO arm of GSE260742, and a mark with no expected relationship to the fusion would be the cleanest negative the panel could have.
  2. Does the two-and-two split hold in the PCGF1-KO genotype? It does for three of five contrasts (H3K79me2 -1.000, H3K4me3 -1.000, H2AK119ub paired +1.000) but H3K27me3 falls to +0.800 and H2AK119ub rep2 to +0.400, and whether that is genotype biology or single-library noise cannot be decided without a replicate.
  3. Would a class-level antibody confound look like this? Nothing in this deposit can answer it, because no antibody is named in any of the 158 sample records.
Every source behind this entry (3)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Adding H3K79me2, the eviction-depth Spearman panel over the same 1,228 S_FLAG bins in murine MLL-AF9 under 500 nM VTP-50469 is H3K27me3 +1.000, H2AK119ub +1.000 in both conventions, H3K4me3 -1.000 and H3K79me2 -1.000, with spans 1.0227, 0.3820, 0.1567, 0.2534 and 0.6005.

    What this evidence cannot show

    NO SPIKE-IN NORMALISATION IS POSSIBLE. Q117 verified zero hits for the entire spike-in vocabulary across all 419,824 characters of GSE260742's 158 sample records and no scale factor is deposited, so every cross-condition number here is a WITHIN-SAMPLE SHARE. It cancels the unknown per-sample scale factor exactly and IS BLIND TO A UNIFORM CHANGE BY CONSTRUCTION: nothing here measures H3K79me2's abundance, everything measures its distribution. EXPOSURE ASYMMETRY: DMSO 72 h and SGC0946 5 uM 72 h are matched, VTP-50469 500 nM is 48 h, so every VTP ratio is a LOWER BOUND on the 72-hour effect - identical in both genotypes, so the genotype DIFFERENCE is unaffected. REPLICATION, CORRECTED MID-BEAT: the NTvKO arm deposits ONE H3K79me2 ChIP per genotype per drug, so ITS spans carry no technical error bar and its bin-level bootstrap intervals measure sampling within one library, NOT library-to-library variation. A REPLICATE PAIR DOES EXIST for a different culture - GSM8123372/374 and GSM8123386/388, both titled MLLAF9 Control, the second carrying `rep2` - already binned by Q122 and scored in Q130 DEVIATION 4: the SIGN and the monotonicity reproduce (Spearman -1.000 in all four murine VTP contrasts in this deposit) while the SPAN reproduces only to within a factor of 1.99 (-0.2438 against -0.4847). EVERY SPAN IN THIS BEAT SHOULD BE READ WITH THAT 2-FOLD REPRODUCIBILITY. THE KNOCKOUT is one CRISPR guide, one clone C1, no second clone, no rescue, and NO deposited PCGF1 protein or transcript measurement. S_FLAG is defined by a DYKDDDDK-tag ChIP on a RETROVIRALLY OVER-EXPRESSED tagged MLLAF9 construct, so it is the over-expressed molecule's sites and not an endogenous fusion's. THE H3K79me2 ANTIBODY IS NOT RESOLVED - `antibod` occurs zero times in the 158 GEO records - and is recorded unresolved rather than guessed. SYSTEM: murine MLL-AF9 (KMT2A::MLLT3) myeloid blood cells, NOT the infant B-lineage KMT2A::AFF1 disease this graph is about. PANEL-SPECIFIC: comparing spans ACROSS marks assumes the marks' dynamic ranges under this normalisation are comparable, which is not established and cannot be from this deposit. The H2AK119ub PAIRED arm is .filt.bam.tdf while the other four contrasts are bare .tdf.

    Evidence that argues against this (2)
    • A published methods study establishes that STANDARD ChIP-seq normalisation FAILS TO DETECT a genome-wide decrease in a histone mark under an inhibitor of its writer: standard methods see no H3K27me3 reduction under an EZH2 inhibitor, and a substantial reduction appears only with a Drosophila chromatin spike-in. Every number in this entry is exactly the kind of normalisation that paper shows to be blind. Egan B et al. 2016
    • The system is murine MLL-AF9 (KMT2A::MLLT3) myeloid cells carrying a retrovirally over-expressed FLAG-tagged construct. This graph exists for human infant B-lineage KMT2A::AFF1, and this corpus's commonest documented synthesis failure is generalising a murine MLL-AF9 myeloid result to that disease. Neville D et al. 2026
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    The three non-H3K79me2 rows of the panel were computed and committed by Q125 in cycle 88 and are read from its deviation2-quartiles-and-conventions.json rather than retyped.

    What this evidence cannot show

    Read in code at emit time, so the panel cannot drift from Q125's committed values without this entry's regeneration changing too. Q125's own caveats apply unchanged, including that the H3K27me3 and H3K4me3 antibodies are unresolved in the GEO records.

    Evidence that argues against this (1)
    • Q125 itself recorded that its H2AK119ub reagent cannot distinguish PRC1.1 from canonical PRC1, so the repressive half of the panel is less mechanistically specific than the Spearman table makes it look. Iqbal Lab oncofusion knowledge graph 2026
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    The four eviction quartiles are genuinely not interchangeable - all three deposited inputs carry a monotone gradient across them at Spearman -1.000, with max/min 1.1507 (NT), 1.1686 (C1) and 1.1297 (timecourse) - and the mark ratios still split two positive and two negative, which no bin-level property can produce.

    What this evidence cannot show

    An input gradient of 1.13-1.17 fold is small against the observed H3K79me2 range of 2.02-fold, and it cancels in a within-genotype drug/vehicle ratio because the input enters neither term. It is quoted here as evidence that the bins DO differ structurally, which strengthens the sign-reversal argument rather than weakening it.

    Evidence that argues against this (1)
Connected to